Meta新研究：EvoHarness-RL让智能体学习协调策略
Meta 新论文让智能体学会管理外部记忆，Qwen3-8B 在 ALFWorld 到 96.9%，长任务不用堆工具。
Meta 新研究提出 EvoHarness-RL，让智能体离线学习 harness 策略并在运行时在线构建外部状态。该策略利用 Belief、Progress、Experience 三类状态，先通过监督微调学习动作空间，再使用成本感知 GRPO 优化。基于 Qwen3-8B 的智能体在 ALFWorld 基准上达到 96.9%。训练中出现 harness annealing 和 harness evolution 两种动态，前者将重复模式吸收进模型策略，后者压缩工作区为紧凑状态。研究认为长程智能体更取决于可训练协调策略，而非更大工具或记忆。
New research from Meta. Agent harnesses are still mostly authored by hand. This makes it hard to tune robust agent harnesses for long-horizon tasks. In this new work, agents learn harness policies offline and deploy them to construct and update external harness state online during runtime task execution. EvoHarness-RL learns that policy instead. Belief, Progress, and Experience are exposed as harness state the policy can act on. Supervised harness fine-tuning teaches the action space, then cost-aware GRPO explores when to read, update, and consolidate during a long run. Qwen3-8B reaches 96.9% on ALFWorld. Two dynamics come out of the training. > Harness annealing means recurring harness-use patterns get absorbed into the model policy, and the agent shifts from frequent calls toward selective access. > Harness evolution means progress updates and experience consolidation compress the workspace into a compact task-adaptive state. This shows that long-horizon agents get more from a trainable coordination policy than from bigger tools or larger memories. Paper: arxiv.org/abs/2608.05446 Track more trending AI papers in our academy: academy.dair.ai 💬 12 🔄 18 ❤️ 131 👀 10673 📊 46 ⚡